Detection And Visualization Study In Early Stage Of Maize Leaf Disease Based On Hyperspectral Technology

While maize growing,pathological changes will occur in morphology,physiology and biochemistry because of the effects of biological or abiotic factors.And it will prevent maize of growing developing and fruiting.In the early stage of maize disease infection,if diseases can be found and judged timely and correctly,and pesticide application rationally,we can not only control the further development of diseases effectively,but also achieve the purpose of reducing environmental pollution by precise application of pesticide.It is of great significance to improve the quality and yield of maize.In this paper,maize leaves were taken as the main research object.Based on hyperspectral technology,maize leaf fungal diseases(Curvularia lunata(Wakker.)Boedijn and Aureobasidium zeae(Narita et Hiratsuka)Dingley)with higher incidence and greater damage were selected in Northeast area of maize production.Through experimental study,we identified maize leaf disease in its early stage and realized three dimensional visualization of maize diseases.The experimental study includes detection in early stage of Curvularia lunata when the disease doesn't appear on the leaves,classification and recognition of leaves infected classification recognition and three dimensional visual simulation of infection process of maize disease.The methods and conclusions provided a new idea for detection in early stage of maize diseases,and established the theoretical foundation for fast,nondestructive and automatic detection of other plant leaf diseases.Main contents and conclusions:(1)Detection in early stage of maize leaf diseases based on spectral characteristics under laboratory conditionsUnder laboratory conditions,detection in early stage of Curvularia lunata based on spectral characteristics was carried out by in vitro inoculation.We monitored disease pathogenesis from time dimension based on histological observation of the infection process of Curvularia lunata and collecting hyperspectral information from leaves at time points.By analyzing the difference of spectral characteristics between inoculated leaves and normal leaves,we used mixed distance method to confirm the feature bands of Curvularia lunata,established disease identification model based on Error BackPropagation(BP)neural network for samples collected at different time points based on feature band and realized detection in early stage of maize leaf diseases in maize.The study results show that 465.1nm,550.7nm,681.4nm were feature bands of Curvularia lunata according to the analysis of spectral differences between inoculated leaves and normal leaves through mixed distance method.The BP neural network model based on the feature band was used to identify the disease.The accuracy of the model was 86.25~92.5% when the inoculated leaves were in the early stage without obvious symptoms through testing samples at different time points.When the green leaf spots on the surface of the inoculated leaves appeared,the accuracy of identification of leaf with spot was 97.5%.(2)Detection in early stage of Curvularia lunata under field conditionsBased on in vitro inoculation experiment in laboratory,the study adopted the live inoculation in the field,in which hyperspectral imaging system and hand-held field spectroradiometer were used to collect spectral data of maize leaves inoculated with Curvularia lunata in incubation period separately,and the spectral characteristics of inoculated and uninoculated leaves were analyzed.The feature bands were selected by correlation analysis and significance analysis.Based on support vector machine(SVM)method was established to detect the disease samples in the early stage.The average spectrum of inoculated leaves was different from that of normal leaves when no symptoms of disease infection were found on the surface of the inoculated leaves.But whether or not the difference is caused by pathogen infection is the key to determine if the spectral difference can be used for detection in early stage of maize leaf disease.In the study,after the hyperspectral data acquisition of the sample leaves,the necessary shaping and transparent treatment of the leaves were carried out,and the leaves were observed by metallographic microscope system.The results of microscopic observation showed that all the maize leaf samples inoculated with Curvularia lunata were observed from germination to mycelium invasion stomata,and all the inoculated maize leaf samples were successfully inoculated by microscopic examination.The spectral data of maize leaf collected by hyperspectral image acquisition system were obtained which can be applied to the early identification of maize diseases.The results show that: the accuracy rate of disease leaf identification reached 79.5~88.75% in the early stage of no obvious symptoms of the inoculated leaves,for the field data collected.(3)Classification and identification of different diseases of maize based on spectral characteristicsCurvularia lunata and aureobasidium zeae are the main leaf diseases of maize in Northeast China.Meanwhile,the two diseases are similar and difficult to distinguish.Based on hyperspectral imaging system,hyperspectral data of Curvularia lunata and Aureobasidium zeae were collected,and spectral characteristics of two kinds of diseased leaves were analyzed.Based on the central limit theorem and law of large numbers,the feature bands region of diseased leaves was determined based on mean confidence interval,and the correlation analysis was used to further extract the more representative feature band.Based on support vector machine(SVM)method was established to classify the disease samples in period of disease.The results show that: the accuracy rate of disease leaf classification reached 96.7% in period of disease.(4)Three dimensional visualization method of maize disease based on imageThe study proposed a three dimensional visualization method of maize disease based on image.The color and morphological features of the spot were extracted from a single image of the spot,and the static features were analyzed,the dynamic process of the spot was deduced.Then the distribution of the spot on maize leaves was set based on user interaction.Finally,a dynamic material model was constructed to simulate the temporal and spatial scale changes of the reflective and transmissison characteristics of the spot.The three dimensional visualization simulation of maize disease infection process was realized.The results show that this method can make the three-dimensional simulation veritably of the apparent changes of maize caused by disease,and provide an effective visualization tool for intelligent decision-making of agricultural diseases,the agricultural science popularization and skill training.